In order to achieve precision measurement of the spring parameters and error comparison, three-dimensional spring multi-parameter measurement system was designed. A three-dimensional spring error comparison theory was established. Algorithm such as pinstripe exponential growth stereo matching technology, pattern recognition method for rotational inertia based on image normalization, linear least square method were applied in this paper. First, high-precision cyclic coded targets were applied in binocular camera calibration, Next, take advantage of structured light stripe technology, it was implemented to stereo matching. Then, Normalized moment of inertia was introduced for feature extraction to obtain a three-dimensional point cloud of the spring. The chord method was used to reduct noise, and the STL model was obtained by filtering. And then the three-dimension length and cylindricity of the spring wre evaluated through Cylindrical fitting and plane fitting of 3D point choud. Finally, a comprehensive spring three-dimensional error comparison theory was set up to comply a rapid identification of unqualified spring. The results show that three-dimensional point cloud detected uniquely associated with the object and the depth of field can reach 300 mm, thus avoiding errors due to the changes in the relative position of the camera and the spring, and the combination of two-dimensional image and three-dimensional detection, so that the accuracy can reach 0.05 mm. The spring measurement system has a faster detection speed due to the comprehensive spring three-dimensional error comparison theory. The proposed three-dimensional visual spring detection method is high-precision, non-contact and non-destructive measurement. The method also has advantage of stable and fast detect speed which meets the requirements of online testing.